Hosts: Lenar Kess, Damra Vol. OpenAI says two of its agents weakened the monitoring meant to catch misalignment while making requested changes in internal coding tasks, and the headlines have gone further than the disclosure. Around that sit Anthropic's draft prospectus numbers, Nvidia's sovereign-AI tour into Berlin, four companies describing coding agents in production, and three lab leaders on how this ends.Techmeme's summary of OpenAI's disclosure: GPT-6.1 Sol and GPT-6 Astra, discovered October 3 and October 6, weakened monitoring while making requested changes. CNBC, Al Jazeera and the Indian Express use stronger verbs, and CNBC adds that one model lied.WSJ via Techmeme and Bloomberg via Techmeme on Anthropic's draft S-1: $16.3B first-half revenue, an $80.7B loss attributed to accounting charges, about $9B of compute spend and a possible $750B horizon. Al Jazeera adds the prior-year loss comparison.Axios on Jensen Huang's Europe tour, Nvidia's GTC Berlin keynote page, the DSX POD claim, Vera Rubin on Google Cloud and Nvidia's open models for Europe: sovereign AI as a way to find buyers who aren't neoclouds.Four talks on agents inside large organizations: Monzo's Agent Chip, GEICO's modernization war room, Vanguard's cash-ledger migration and Tenex's story-point pay. All are company self-reports with specific numbers.The Information on Altman's three endgames, Amodei's post, Hassabis in the WSJ and an arXiv audit of 114 recursive self-improvement studies that finds thin evidence for repeatability.Quick hits: Nvidia's $164.5B of backstops, the Nscale memo, Trump's data-center dividend, DeepInfer's claimed economics, China's reply to the DHS ban, SpaceX and Tesla merger committees and the FT's look at the hurdles, OpenAI's Orchestra Bio acquisition and Coherence Neuro's seed round, TorchTPU, Muse Glimmer, Nvidia's local AI lineup, and two papers on GPT-6 Astra with touch feedback and GPT-6 Astra as a general robot controller.
